Machine learning prediction models based on hub genes related to immune phenotypes in muscle-invasive bladder cancer treated with immune checkpoint blockade

Abstract Background Bladder cancer is one of the most frequent cancers in the world. Muscle-invasive bladder cancer (MIBC) is the aggressive subtype with higher morbidity and mortality. Immune check point blockade (ICB) therapy has shown its potential for treating MIBC, but is limited due to the lack of predictive biomarkers. Methods 1601 MIBC transcriptomic profiles were obtained from 10 datasets. Unsupervised clustering of immune phenotypes in MIBC was performed based on immune-related signature genes selected by us. We analyzed the characteristics including microenvironments, metabolic pathways, and survival rates in different phenotypes. Multi-omics analysis and WGCNA plus protein-protein interaction (PPI) analysis were performed to identify hub genes of differentially expressed genes (DEGs) distinguishing phenotypes related to prognosis. The hub DEGs were further validated by real-time quantitative PCR (qPCR). A model was established and CART was employed to predict the responses of patients treated with ICB. Results Of various immune phenotypes, cluster 3C was the most “inflamed” subcluster with the best prognosis, while cluster 1A was associated with “non-inflammation” and worst prognosis. There was no intersection of hub DEGs selected by WGCNA plus PPI analysis and multi-omics analysis. WGCNA plus PPI analysis identified 5 hub genes related to the survival rate of patients, IFNG, CXCR6, IL2RB, LCK, and PSMB10. Real-time qPCR results indicated that the expression levels of 5 hub genes were significantly lower in tumors. The 5 hub genes were further utilized for prognostic score model and decision tree analysis. The areas under the curve (AUC) of the ROC curves predicting 5-year endpoint generated from the risk-based prediction model were 0.652. The mean accuracy, sensitivity, and specificity of CART for predicting stable disease/progressive disease were 70.1%, 70.0% and 71.7%. Conclusions The 5 hub genes and generated models showed the potential for predicting the prognosis for patients receiving ICB therapy. The molecular mechanisms regulating the expression of the hub genes require further studies in the future.